The way around this is to turn on the surface from the beginning of the track so you can see if the track is still following. If the plane is clear, going back a few frames and adjusting the spline to accommodate a slip and then retracking the problem frames should fix the problem in most instances, rather than adjusting the spline at the point where you see the problem.
Constantly adjusting the spline at the point of slipping is good if you’re just doing roto, but if you need it for match-moving or inserts, the surface and grid are going to be your friends so that you can fix the problem before it compounds further down the timeline.
On the export side, it’s not so much a matter of averaging the spline points, because splines that have been manually adjusted to follow an object are not going to be realistically representing the tracked position of the plane in question. You may be able to get a general locked position on a single point of reference, but it won’t accurately define a 4-point plane.
Having said that, it would be interesting to see if we can make such an export viable for other tasks. We’ll look into it.